Re: [Sbcl-devel] Killing runaway threads

[<me...@ho...>]

On Sunday 18 June 2006 15:35, Nikodemus Siivola wrote:
> Currently there is no way in SBCL to kill runaway threads without
> killing the whole image.
>
> Example:
>
>  (defun bad ()
>    (unwind-protect
>        (progn
>          (write-line "/i'm a little runaway!")
>          (loop))
>     (bad)))
>
>  (defvar *bad* (sb-thread:make-thread #'bad :name "Bad"))
>
> ...now good luck killing *BAD*.
>
> Here's a first cut in making this possible. First cut not in the
> sense that I think there is anything wrong with it, but in the sense
> that I haven't enough time on thinking about the interface. More of
> that in another post.
>
> All that happens here is that the thread exit related cleanups have
> been moved into HANDLE-THREAD-EXIT and handle_thread_exit on the
> C-side, and instead of returning we always end up calling
> pthread_exit.
>
> Does the implementation look reasonable to the more thread-savvy
> among us?

The patch looks reasonable to me. About the interface, instead of=20
THREAD-EXIT (which should be EXIT-THREAD anyway), you could undeprecate=20
DESTROY-THREAD and make it interrupt the target thread with (lambda ()=20
(handle-thread-exit thread)).

I don't really like that SB-EXT:QUIT either exits the current thread or=20
the whole image depending on the current thread being the initial=20
thread, but it's understandable for backward compatibility reasons. So=20
if the contract of QUIT is to stay the same, one could hook up=20
destroy-thread to the recklessly-p branch, killing=20
the '(sb-thread:make-thread (lamba () (quit :recklessly-p t))) exits=20
the image' bug.

Cheers, G=E1bor